Rasearch On Parameter Optimization Of Support Vector Machine Based On Pigeon-inspired Optimizaton Algorithm

Pigeon-inspired optimization(PIO)is a novel group intelligence algorithm proposed by duanhaibin in 2014.The algorithm simulates the behavior of the pigeons' cooperative homing to solve the optimization problem.There are two independent operators in this algorithm,the map and compass operators mainly realize the global optimization of the flock of pigeons,and the landmark operator is used to realize the rapid convergence of the flock of pigeons.In addition,the principle of this algorithm is relatively simple,requires fewer parameters,and is easy to implement.After being proposed,this algorithm has attracted more researchers' attention.At present,this algorithm has been applied to such problems as scheduling,image segmentation,path planning and so on.Although the flock optimization algorithm has good performance,when solving the higher dimension optimization problem,there are often algorithm convergence accuracy is low,it is difficult to escape the constraint of the local optimal solution.In order to solve these problems of the flock optimization algorithm,this paper proposes an improved flock optimization algorithm based on the study of the principle of the flock optimization algorithm,and applies the improved flock optimization algorithm to Support Vector Machine(SVM)parameter optimization,and to realize the image classification research.The main tasks are as follows:Firstly,the greedy update strategy is adopted in the process of individual status update.In the later stage of the algorithm,the diversity of the flock is easily reduced faster.In order to improve the ability of the flock optimization algorithm to jump out of the local extreme constraint,a state update rule of simulated annealing is proposed to give the poor state a chance to be accepted.At the same time,the inertia weight of linear decline is added to the map and compass operators,so that the early search in a large range can prevent the emergence of the local optimal situation,and the later local search can make the algorithm fast convergence.At the later stage of the algorithm,the population communication behavior is increased to improve the algorithm's diffusivity.Secondly,support vector machine can solve the classification problem well.However,the selection of kernel parameters in SVM determines the performance of SVM,and there is no rule to follow in setting the parameters.In this paper,the mechanism of parameteroptimization of support vector machine(SVM)based on flock optimization algorithm is studied.Finally,an improved support vector machine(SVM)for image classification is presented.In this process,SIFT features of the image are first extracted,and then the SIFT features are organized based on the word bag model and histogram.Then,SIFT features after organization are combined with kernel functions of support vector machine to construct different image classification systems.Experiments show that the proposed method is effective for image classification.